Head-to-head · updated 13 September 2026

DataShield vs Amundsen: where do you take a self-hosted data catalog that just got archived?

Amundsen was a good idea, done early and given away. Lyft open-sourced it in 2019, the Linux Foundation took it in 2020, and for a while it was the default answer to "we need a search box over our tables". The popularity ranking was clever. The install was light. Thousands of teams starred it and a fair number ran it.

It is over now. LF AI & Data voted to archive the project in September 2026, the GitHub repo carries an archive banner, and the last real release shipped in August 2024. So this page is not a feature duel. It is a migration read for the people still running it. DataShield Ontology is a self-hosted catalog with field classification, stewardship and an agent surface built in. Datasets are tokenized at ingest; agents query tokenized data over MCP; detokenization is a privileged, audited operation.

Get free Demo Center access Get an instant quote

The short version

Pick DataShield when

You want a catalog someone is still shipping. Ours gets releases. Amundsen got an archive banner on 10 September 2026.
You need the catalog to know what the data is, not just where it lives. We label columns against 129 PII, PHI, financial and secret classes, and the verdict is reproducible from a config digest.
Agents read your tables now. We expose the catalog over MCP, gate every governed call, and re-check authority mid-session. How Auth does it.
You have to prove who read what. Our audit log is a hash chain with signed checkpoints, and erasure does not break it. Run the verifier.

Pick Amundsen when

It already works and nobody is asking it to do more. Archived code does not stop running. If your Neo4j and Elasticsearch boxes are quiet, you can leave them quiet.
You need Databuilder's extractor list today. Hive, Presto, Redshift, Snowflake, BigQuery, dbt, Airflow. Our provider scanning is PostgreSQL only right now.
Cost has to be zero. Apache-2.0, no seats, no renewal, and nobody left to raise the price.
You want a search-first analyst UI with popularity ranking. That was the best thing Amundsen did and we do not copy it.

Bottom line: Amundsen answered "where is the table". The question now is "who may read this column, and can you prove they did". That is a different product, and it is the one we build.

Feature by feature

Competitor cells describe what Amundsen's repo, releases and LF AI & Data project page say as of the date above. If we've mischaracterised something, email support@myorg.ai and we'll correct it, credited.

What matters	DataShield	Amundsen	Edge
Project status	Live in production (Guardian and Lighthouse since April 2026). Small team, active releases, SOC 2 not yet certified and we say so.	Archived. The repo banner reads: "Due to inactivity, this project was archived in September 2026." LF AI & Data's page says its board voted to archive. Last release August 2024.	◆
Catalog and discovery	Register a provider over a live connection, scan it, and profile assets and columns in place with no row egress. PostgreSQL is GA today. Snowflake, BigQuery, Databricks, S3 and Salesforce are declared and not built yet.	Databuilder shipped extractors for a long list of warehouses, BI tools and schedulers. Broader than us on day one, frozen at the 2024 version of each.	◇
Classification	129 built-in field classes, including all 18 HIPAA Safe Harbor identifiers and eight non-US national ID formats. Regex plus checksum validation, column-name lexicons and anti-pattern suppressors. No model, so no drift. It cut over to live days ago, so call it shipped rather than proven.	No PII or PHI classification. Tags and descriptions are whatever a human typed in.	◆
Sample values in the catalog	Default policy is `omit`: no column value is copied into the catalogue. Storing plaintext samples needs an explicit acknowledgement and writes an audit row.	We couldn't find a sample-value policy in what remains of the public docs. The documentation site now redirects off the project entirely, so ask whoever runs your instance.	◆
Lineage	Typed lineage traversal with access gating at every hop, so an edge never reveals an asset you can't read. It is derived from the pipelines that own the relationships, not a stored column-level graph.	Table-level lineage where an extractor supplied it. No per-hop access gating, because the graph has no opinion on who is asking.	◈
Audit evidence	SHA-256 hash chain with Ed25519-signed checkpoints that are themselves chained. Verification names the failure: tampering, insertion, deletion or truncation. Browser demo.	Application logs. Nothing that survives someone with database access and a bad week.	◆
Agent authorization	Every governed tool call passes a scope ceiling, a declared authority tier and a revocation re-check before dispatch. Revocation lands mid-session.	No actor model beyond a login for the frontend. Nothing to authorize an agent with.	◆
Break-glass	Scoped, time-boxed emergency access for agents. It auto-revokes and can't be quietly deleted from the log.	Not a concept in the project.	◆
GDPR erasure	Crypto-shred of per-subject key material plus ISO 27560 consent receipts. The chain still verifies after erasure.	A metadata catalog holds metadata, so erasure is a question for the systems underneath it.	◆
Tokenization and data handling	Deterministic, join-preserving, vault-reversible tokens applied at ingest, plus quasi-identifier generalization (dates to year, decade or age band; ZIPs to 3 or 4 digits; partial phones, SSNs and emails) with a measured cardinality-reduction score per column. Masking and generalization are switches you turn on, not defaults.	None. Amundsen never touched the data, only the metadata about it.	◆
MCP and agents	More than 200 MCP tools across Ontology, Auth, Corpus and Lighthouse. An agent can search the catalog, read a profile, and act on it under a tool token with a scope ceiling.	No MCP server and no agent story. The project stopped shipping before the question was asked.	◆
Deployment	Self-hosted in your own cloud or data center, or a dedicated single-tenant server we operate. Docker images for Auth, Ontology, Corpus and Lighthouse, with a signed deploy manifest.	Self-hosted only. Neo4j or Atlas for the graph, Elasticsearch for search, plus frontend, metadata and search services. No managed option exists now that Stemma's is gone.	◈
Pricing	Published model, scoped instant quote, no sales wall.	Free forever, Apache-2.0. We are not going to pretend to win this row.	◇

◆ DataShield leads◇ Amundsen leads◈ comparable

Amundsen claims are drawn from github.com/amundsen-io/amundsen, its releases page, and the LF AI & Data project page, last checked 13 September 2026. Links are below rather than paraphrased from memory.

Three things you get here that you won't get from a metadata catalog

The catalog knows what the column holds

Amundsen told you a column was called mbr_dob and that 40 people queried it last month. Useful. It did not tell you the column was a date of birth under HIPAA Safe Harbor. We label it, cite the rule, and stamp the verdict with a config digest so you can re-derive the same answer next quarter. What Ontology does.

Proof that survives an audit

A log you can edit proves nothing. Ours is a hash chain with signed checkpoints, and the verifier says what broke, not just that something did. That is the property EU AI Act Article 12 and HIPAA §164.312(b) reviewers look for. Try the demo in your browser, no signup.

Authority that can change mid-flight

An analyst leaves on a Friday. Their agent is still 20 minutes into a job over your tables. We re-check authority on the next governed tool call and fail closed. A search index has no view on this, because it doesn't know who is asking. How Auth does it.

Where Amundsen is genuinely stronger

Credit where it is due. Amundsen shaped the modern catalog category. It was the first tool many teams used that treated data discovery as a search problem, and ranking tables by how often people actually queried them was a smarter default than anything the enterprise suites shipped at the time. Over 200 people contributed. It cost nothing, it installed in an afternoon, and Lyft never tried to sell it back to you. Databuilder's extractor list still beats ours across a mixed warehouse estate, and that gap is real: we scan PostgreSQL in place today and everything else is on a roadmap, not in a release. If you run Snowflake and Redshift and want a table search box this week, we are not the short path.

Here is the push-back. Free stops being free when nobody ships a patch. The docs site now redirects to an unrelated domain, the maintainers merged a commit in August 2026 to strip their own broken documentation links, and a month later the board archived the repo. Your CVE response plan for Neo4j and Elasticsearch is now your own problem, and so is every dependency under them. Teams in this spot usually pick one of three doors: fork it and own it, move to another open-source catalog, or ask whether the catalog should have been doing more than search all along. We are behind door three, and we would rather you walked through it on purpose than because something broke.

Questions worth asking both of us

These are the questions we'd want answered if we were buying. Ask them on every vendor call, including ours.

Can you cryptographically prove an audit log entry wasn't deleted?

DataShield: yes. Each record commits to the one before it, checkpoints are signed and chained, and verification tells deletion apart from truncation and tampering. Try it at /verify. Amundsen: no. It logs like any Flask app, into a store you can edit. That was fine for a discovery tool and it is not enough for an examiner.

What happens to a revoked agent mid-session?

DataShield re-checks authority on every governed tool call, so revocation bites on the next call. Amundsen has no agent model at all, so the honest answer is that the question doesn't apply, which is a different thing from a good outcome.

How does GDPR erasure interact with the audit trail?

DataShield crypto-shreds per-subject key material and issues an ISO 27560 consent receipt. Actor identities in the chain are HMAC-committed, so the evidence still verifies after the subject is gone. Amundsen stores metadata, not subjects, so erasure lands on the warehouse behind it and on whatever you built to track it.

Amundsen is archived. Can we just keep running it?

Yes, for a while. Archived code keeps working and the images still pull. What stops is patches. You now own the security posture of Neo4j or Atlas, Elasticsearch, and every Python dependency in Databuilder, with no upstream to accept your fix. The public documentation moved off the project's own domain, so onboarding a new engineer means reading source. Plan a horizon, not a panic.

Are you a drop-in replacement for Databuilder's connectors?

No, and anyone who says otherwise is selling. Our catalog provider scanning is GA for PostgreSQL. Snowflake, BigQuery, Databricks, SQL Server, MySQL, Salesforce, Kafka and object stores are declared with no handler yet. Separately we crawl open-data portals and watch S3, Azure Blob, GCS, SFTP and SMB for files to ingest. If your estate is mostly Postgres plus files, the swap works today. If it is a Snowflake shop, talk to us about timing before you rip anything out.

Does DataShield have SOC 2?

Not yet, and we won't imply otherwise. Auth is live with a public threat model and a verifier you can run in a browser. Guardian and Lighthouse have been in production since April 2026. Design-partner terms include source escrow, so a small vendor isn't a single point of failure. Details on the security page.

Amundsen's GitHub repo is archived. Banner: "Due to inactivity, this project was archived in September 2026. The contents will remain available for historical purposes." 4,785 stars, 961 forks, 0 open issues. — GitHub, amundsen-io/amundsen, 13 Sep 2026
"The LF AI & Data TAC and Governing Board voted to Archive the Amundsen project in September 2026." Contributed by Lyft in July 2020 as an incubation-stage project. — LF AI & Data Foundation, Amundsen project page, 13 Sep 2026
Last release is databuilder-7.5.1, published 14 August 2024. The frontend, metadata, search and common components last shipped 6 March 2024. — GitHub releases, amundsen-io/amundsen, 13 Sep 2026
PyPI confirms amundsen-databuilder 7.5.1 as the current version, uploaded 14 August 2024. — PyPI, amundsen-databuilder, 13 Sep 2026
Maintainers merged "fix: Removed links in README to amundsen.io" on 20 August 2026. The project domain and its GitHub Pages docs mirror now redirect to an unrelated site. — GitHub PR #2365, amundsen-io/amundsen, 13 Sep 2026
EDPB names pseudonymization as a GDPR risk mitigation for AI models. — EDPB Opinion 28/2024

Other head-to-heads

Same market

Get free Demo Center access

You've seen the proof

Ready for a number? Scope your deployment and we'll price it against your own economics.

Get your quote →